Rust implementation of Andrej Karpathy's makemore YouTube series
makesmores
takes a database of strings (in this case the 32k most popular baby names from ssa.gov for the year 2018) and generates more of them.
Currently allows for the choice between two different model architectures:
- MLP, following Bengio et al. 2003
- CNN, following DeepMind WaveNet 2016
This is intended as a low-level exploration of neural networks, so although it depends on torch
, it only utilizies low level Tensor
operations. Layers of the network are implemented by hand. Hyper parameters can be adjusted via the constants at the top of the file.
Although this code represents the product at the end of the video series, for those following along, the partial implementations can be found in the commit history.